Q&A: Machine vision software, advancing algorithms

The following is a discussion with Steven Gentner, founder of RoboRealm.

Name: Steven Gentner
Company: RoboRealm
Title: Founder

What is your latest product and for what reason was it developed?

RoboRealm specializes in machine vision software and develops a desktop-based machine vision application for use in industry, academic, and hobbyist markets. The application was developed to create an easy way to process and analyze images from various sources on a generic computing device in an interactive way.

We found that being able to configure processing parameters live while viewing the results was very effective in helping users understand the function of different processing methods rather than learn having to learn the mathematical or algorithmic background of the techniques!

What is your take on the current state of the machine vision market?

We see the current machine vision market as being very strong and expanding. Within industrial markets, we see more vision being used as more companies look towards adding automation within their production processes or in enhancing existing systems to check more aspects of their products.

In academic markets there is a clear strategy on exposing students to machine vision technologies because of increased emphasis on STEM initiatives which typically include vision-guided robotics as a learning activity. Finally, within the hobbyist community, vision is a key area of interest within embedded devices for use on various platforms and has had a couple new products introduced into this space with good success.

Are market changes affecting your product development, and if so, how?

We react to market changes through our customers. As a customer-driven company we rely on customer requests to help prioritize development. This happens across all markets and helps us to remain current with recent market changes but only those that are relevant to our customers.

As it’s very difficult to forecast acceptance towards any new technology, its best to first ensure customer interest before committing resources. Having customer reactions also helps us understand how new technology would be best utilized.

In which areas or applications do you see the most growth?

We see growth “across the board” as technologies improve and become more cost effective for all markets. Within industrial markets there is a push for larger images and faster processing using parallel technologies such as CUDA or OpenMP. Larger images allow for more detailed inspection of more products at once but require faster processing in order to retain a desired throughput.

The academic and hobbyist markets are more price and size (form factor) sensitive which are well accommodated by the increasing computing power in smaller cheaper platforms such as the Raspberry PI, among many other new offerings.

Is there a particular trend or product in the next few years that you see as “the next big thing?”

We had predicted and anticipated 10 years ago that computing would continue to increase in speed at lower cost. While that's not a very difficult prediction to make, it is key to vision processing that relies heavily on computing power.

For the average consumer computing power is probably well beyond their needs but for vision processing is easy to tax even the fastest machines with capabilities that humans take for granted. More computing power, faster data communications (i.e. leveraging the cloud), and much more R&D on vision algorithms is already becoming the “next big thing.”

How will imaging systems have to change to meet emerging applications?

Complete imaging solutions are still complicated to implement. Despite all the decisions that need to be made in terms of camera, lens, lighting, computing, etc. that are specific to imaging; there are many more decisions that need to be made in terms of integration with external devices such as PLCs.

The industry constantly works to reduce the complexity of these solutions but for many companies it is still too daunting (or expensive) a task to implement. Even when systems are well-integrated, there is still difficulty in determining what part of the system is responsible for any occasional issues that arise. Some companies have figured this out and created full solutions where software and hardware are created as a single solution (think Baxter) but those solutions are still unique in the industry.

Do you have any new exciting products or developments on the horizon?

RoboRealm will be focusing more on including neural net and genetic algorithms to augment image analysis in the coming year. Too much time is being spent on configuration of machine vision systems that only grows as more complex tasks are attempted. It becomes clear that training or evolving a system is desirable as that can be accomplished by many more people and have a wider appeal to a less technical audience.

Flexibility in configuration is required for most industrial applications, but this flexibility comes at an ease-of-use cost. There are many vision techniques that are not used due to a lack of understanding how the configuration needs to be applied. Self-learning systems would help reduce the complexity of solutions and focus more on good training data as opposed to configuration knowledge. This strategy is mirrored in many of the autonomous driving systems currently being announced by many companies.

Page 1 | Page 2